Each brain imaging modality reports on a different aspect of the brain (e.g. gray matter integrity, blood flow changes, white matter integrity) and each has strengths and weaknesses. However the vast majority of studies analyzes each imaging modality separately and interprets the results independently of one another. Many mental illnesses, such as schizophrenia, bipolar disorder, depression, and others, currently lack definitive biological markers and rely primarily on symptom assessments for diagnosis. One area which can benefit greatly from the combination of multimodal data is the study of schizophrenia. The brain imaging findings in schizophrenia are widespread and heterogeneous and have limited replicability. We show evidence that, in part, the lack of consistent findings is because most models do not combine imaging modalities in an integrated manner and miss important changes which are partially detected by each modality separately. Combining multimodal imaging data is not easy since, among other reasons, the combination of multiple data sets consisting of thousands of voxels or time points yields a very high dimensional problem, requiring appropriate data reduction strategies. There are two important areas that we focus on in this new phase of the project. First, we will focus on developing data fusion strategies that will leverage our initial success in developing ICA-based tools for combining multiple tasks and modalities. We will develop and validate approaches which can scale easily from one to many different data types. In the first funding period we focused mainly on pair-wise combinations of multimodal data. However, the results have convinced us that allowing higher order relationships is also important (e.g. we show pilot data in which using 3-way relationships improves our ability to discriminate schizophrenia and control groups). In this proposal we will significantly expand this work and develop novel methods to efficiently exploit high-order joint information not just pair-wise. Next, we will develop new tools that will identify correspondences among modalities. For example we show that structural and functional patterns of covariation are in some cases remarkably similar to one another and in others cases quite distinct and these relationships can predict diagnosis. We will thoroughly test our approach using a well characterized data set involving multiple illnesses that have overlapping symptoms and which can sometimes be misdiagnosed and treated with the wrong medications for months or years (schizophrenia, bipolar disorder, and unipolar depression). As before, we will provide open source tools and release data throughout the duration of the project via a web portal and the NITRIC repository, hence enabling other investigators to compare their own methods with our own as well as to apply them to a large variety of brain disorders. 36 PUBLIC HEALTH RELEVANCE: Each brain imaging modality reports on a different aspect of the brain with different strengths and weaknesses and there are now literally thousands of putative imaging biomarkers. This project will develop multivariate methods which use higher order statistics to combine diverse information in a scalable manner, identify correspondence among data types and also provide a sophisticated data sharing and management system.